In [1]:
import itertools
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.ticker import NullFormatter
import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
In [2]:
df = pd.read_csv('../../preprocessing/apartments_preprocessed.csv')
In [3]:
df.head()
Out[3]:
title bathrooms bedrooms price square_meters cityname state latitude longitude source luxury
0 Studio apartment 2nd St NE, Uhland Terrace NE,... 1.0 0.0 790 9.393 Washington DC 38.9057 -76.9861 RentLingo 0
1 Studio apartment 814 Schutte Road 1.0 1.0 425 9.858 Evansville IN 37.9680 -87.6621 RentLingo 0
2 Studio apartment N Scott St, 14th St N, Arling... 1.0 0.0 1390 9.951 Arlington VA 38.8910 -77.0816 RentLingo 0
3 Studio apartment 1717 12th Ave 1.0 0.0 925 10.788 Seattle WA 47.6160 -122.3275 RentLingo 0
4 Studio apartment Washington Blvd, N Cleveland ... 1.0 0.0 880 11.625 Arlington VA 38.8738 -77.1055 RentLingo 0
In [4]:
from geopy.geocoders import Nominatim

address = 'United States'

geolocator = Nominatim(user_agent="to_explorer")
location = geolocator.geocode(address)
latitude1 = location.latitude
longitude1 = location.longitude
print('The geograpical coordinate of USA are {}, {}.'.format(latitude1, longitude1))

The geograpical coordinate of USA are 39.7837304, -100.445882.
In [5]:
import folium
# create map of USA using latitude and longitude values
map_usa = folium.Map(location=[latitude1, longitude1], zoom_start=4)

# add markers to map
for lat, lng, propname in zip(df['latitude'], df['longitude'], df['title']):
    label = '{}'.format(propname)
    label = folium.Popup(label, parse_html=True)
    folium.CircleMarker(
        [lat, lng],
        radius=5,
        popup=label,
        color='red',
        fill=True,
        fill_color='#FF5733',
        fill_opacity=0.7,
        parse_html=False).add_to(map_usa)  
    
map_usa
Out[5]:
Make this Notebook Trusted to load map: File -> Trust Notebook

Zelim da ova sekcija odgovori na pitanje: Imam x para, zelim veliki/mali stan, ja sam aktivan/pasivan covek, gde mogu da zivim?
Zelim doduse da dodam posebnu kolonu koja ce da odredi da li je drzava za aktivne ili za pasivne ljude.
https://www.cnbc.com/2022/05/25/these-are-the-best-and-worst-states-to-live-in-for-millennials-in-2022.html
https://www.cnbc.com/2023/01/18/best-states-to-raise-a-family-in-the-united-states.html
https://www.travelandleisure.com/naturally-beautiful-states-in-the-country-6543572

In [6]:
df['turbulent'] = 0
df['family-friendly'] = 0
df['nature-lovely'] = 0

# washington dc, massachusets, utah, illinois, minnessota, oregon, wisconsin, virginia, colombia
turbulent_class = ['WA','DC','MA','UT','IL','MN', 'OR', 'WI', 'VA','CO', 'NY', 'CT', 'NE']

#massachusetts,minnesota, north dakota, vermont, new hampshire, new jersey, iowa, washington
family_friendly_class = ['MA','MN','ND','VT','NH','NJ', 'NE','IA', 'CT', 'WA']

#california, alaska, utah, colorado, hawaii, arizona, washington, wyoming, massachusetts, florida, oregon, montana, north carolina
nature_lovely_class = ['CA', 'AK', 'UT', 'CO', 'HI', 'AZ', 'WA', 'WY', 'MA', 'FL', 'OR', 'MT', 'NC']


df.loc[df['state'].isin(turbulent_class), 'turbulent'] = 1
df.loc[df['state'].isin(family_friendly_class), 'family-friendly'] = 1
df.loc[df['state'].isin(nature_lovely_class), 'nature-lovely'] = 1
In [7]:
X = pd.DataFrame(data=df, columns=['price','square_meters','luxury','turbulent','family-friendly','nature-lovely','longitude','latitude'])
X
feature_names = ['price','square_meters','luxury','turbulent','family-friendly','nature-lovely','longitude','latitude']
In [8]:
from sklearn.preprocessing import MinMaxScaler
In [9]:
scaler = MinMaxScaler()
X = pd.DataFrame(scaler.fit_transform(X), columns=feature_names)
X
Out[9]:
price square_meters luxury turbulent family-friendly nature-lovely longitude latitude
0 0.023790 0.000000 0.0 1.0 0.0 0.0 0.922641 0.436714
1 0.009073 0.000446 0.0 0.0 0.0 0.0 0.801088 0.413434
2 0.047984 0.000535 0.0 1.0 0.0 0.0 0.921553 0.436349
3 0.029234 0.001337 0.0 1.0 1.0 1.0 0.406403 0.652966
4 0.027419 0.002140 0.0 1.0 0.0 0.0 0.921281 0.435922
... ... ... ... ... ... ... ... ...
9917 0.233871 0.552643 0.0 1.0 1.0 0.0 0.736632 0.585536
9918 1.000000 0.768031 0.0 0.0 0.0 1.0 0.437081 0.325673
9919 0.435484 1.000000 0.0 0.0 0.0 0.0 0.919740 0.439768
9920 0.185081 0.059057 1.0 1.0 0.0 0.0 0.956780 0.483039
9921 0.032621 0.069448 1.0 0.0 0.0 0.0 0.842745 0.315099

9922 rows × 8 columns

In [10]:
from sklearn.metrics import silhouette_score

silhouette_scores = []
sse_scores = []

for k in range(2, 11):
    kmeans = KMeans(n_clusters=k, n_init='auto')
    kmeans.fit(X)
    
    silhouette_score_k = silhouette_score(X, kmeans.labels_)
    silhouette_scores.append(silhouette_score_k)
    
    sse_scores.append(kmeans.inertia_)  # Append SSE score

# Create a single plot with two subplots
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 6))

# Plot Silhouette Scores
ax1.plot(range(2, 11), silhouette_scores, marker='o')
ax1.set_xlabel('Number of clusters')
ax1.set_ylabel('Silhouette Score')
ax1.set_title('Silhouette Score for Optimal k')

# Plot SSE Scores
ax2.plot(range(2, 11), sse_scores, marker='o', color='r')
ax2.set_xlabel('Number of clusters')
ax2.set_ylabel('SSE')
ax2.set_title('Sum of Squared Errors (SSE) for Optimal k')

plt.tight_layout()
plt.show()
No description has been provided for this image

Sto je vise klastera, veci je silhouette score, ali to moze da rezultuje u klasterima koji ne sadrze znacajne informacije, tako da cemo se zaustaviti na 6 klastera.

In [11]:
kclusters = 6
kmeans = KMeans(n_clusters=kclusters, n_init='auto')
kmeans.fit(X)
Out[11]:
KMeans(n_clusters=6, n_init='auto')
In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.
KMeans(n_clusters=6, n_init='auto')
In [12]:
from matplotlib import colors

# Kreiranje mape
map_clusters = folium.Map(location=[latitude1, longitude1], zoom_start=4)

# Postavljanje boje za klaster
x = np.arange(kclusters)
ys = [i + x + (i*x)**2 for i in range(kclusters)]
colors_array = plt.cm.rainbow(np.linspace(0, 1, len(ys)))
rainbow = [colors.rgb2hex(i) for i in colors_array]

cluster_counts = [list(kmeans.labels_).count(label) for label in range(kclusters)]  # Broj instanci po klasterima

# Dodavanje legendi
legend_html = """
<div style="position: fixed; 
     bottom: 50px; left: 40px; width: 150px; height: 200px; 
     border:2px solid grey; z-index:9999; font-size:14px;
     background-color:white;
     ">&nbsp; Cluster Legend <br>
     """

for i, color, count in zip(range(kclusters), rainbow, cluster_counts):
    legend_html += f"<i class='fa fa-circle' style='color:{color}'></i> Cluster {i}: {count}<br>"

legend_html += "</div>"

map_clusters.get_root().html.add_child(folium.Element(legend_html))

# Dodavanje markera na mapu
for lat, lon, poi, cluster_label in zip(df['latitude'], df['longitude'], df['title'], kmeans.labels_):
    label = folium.Popup(str(poi), parse_html=True)
    cluster_color = rainbow[cluster_label]
    folium.CircleMarker(
        [lat, lon],
        radius=5,
        popup=label,
        color=cluster_color,
        fill=True,
        fill_color=cluster_color,
        fill_opacity=0.7).add_to(map_clusters)

map_clusters
Out[12]:
Make this Notebook Trusted to load map: File -> Trust Notebook

Graficki prikaz klastera¶

In [13]:
pd.set_option('display.max_columns', None)
In [31]:
import matplotlib.pyplot as plt
import pandas as pd


# Podešavanje broja klastera
kclusters = 6

# Kreiranje glavnog platna sa potplotovima
fig, axes = plt.subplots(2, 3, figsize=(18, 10))
fig.suptitle('Cluster Analysis')

# Iteracija kroz klastera
for i in range(kclusters):
    Cluster_i = df[kmeans.labels_ == i]
    
    # Broj instanci sa turbulent = 1, family-friendly = 1 i nature-lovely = 1
    turbulent_count = Cluster_i['turbulent'].sum()
    family_friendly_count = Cluster_i['family-friendly'].sum()
    nature_lovely_count = Cluster_i['nature-lovely'].sum()
    
    # Odabir potplotova
    row = i // 3
    col = i % 3
    ax = axes[row, col]
    
    # Kreiranje grafa
    ax.set_title('Cluster {}'.format(i))
    ax.bar(['Turbulent', 'Family-Friendly', 'Nature-Lovely'],
           [turbulent_count, family_friendly_count, nature_lovely_count],
           color=['blue', 'green', 'orange'])
    
    # Prikaz prosečnih vrednosti kao tekst u gornjem desnom uglu
    avg_price = Cluster_i['price'].mean()
    avg_square_meters = Cluster_i['square_meters'].mean()
    
    box_text = f"Avg price: {avg_price:.2f}\n" \
                f"Avg squarem: {avg_square_meters:.2f}\n"
    
    ax.text(2.5, max([turbulent_count, family_friendly_count, nature_lovely_count]), box_text, va='top', ha='left', bbox=dict(facecolor='white', edgecolor='black'))
    
    # Podešavanje oznaka osa
    ax.set_xlabel('Features')
    ax.set_ylabel('Counts')
    ax.tick_params(axis='both', which='major', labelsize=8) 

# Prikaz grafova
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
plt.show()
No description has been provided for this image
In [26]:
Cluster_2 = df[kmeans.labels_ == 2]
Cluster_3 = df[kmeans.labels_ == 3]

luxury_2 = Cluster_2['luxury'].sum()
luxury_3 = Cluster_3['luxury'].sum()
In [37]:
# da bismo videli razliku izmedju drugog i treceg klastera

# Kreiranje plota i subplotova
fig, ax = plt.subplots()

# Naslov
ax.set_title('Luxury')

# Barovi
ax.bar(['Cluster 2', 'Cluster 3'], [luxury_2, luxury_3])

# Vizuelizacija
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
plt.show()
No description has been provided for this image

Analiza dobijenih rezultata¶

Klaster 0 - najjeftiniji i najmanji¶

Kardinalnost: 2455
Teritorija: Novi Meksiko, Teksas, Luizijana, Oklahoma, Arkansas, Misisipi, Alabama, Dzordzija, Juzna Karolina, Kansas, Misuri, Kentaki, Tenesi, Indijana, Ohajo, Micigen, Pensilvanija
Dominanta kategorija: /
Prosecna cena: 1120 dolara
Prosecna kvadratura: oko 80 kvadrata

Klaster 1 - najskuplji i najveci¶

Kardinalnost: 1921
Teritorija: Severna Karolina, Florida, Arizona, Kalifornija, Montana, Aljaska, Havaji
Dominanta kategorija: Nature-lovely skoro u celosti
Prosecna cena: oko 2050 dolara
Prosecna kvadratura: oko 95 kvadrata

Klaster 2 - za svakog po nesto, luksuzni¶

Kardinalnost: 1192
Teritorija: Nebraska, Ajova, Minesota, Njujork, Njudzersi, Konektikat, Masacusets, Vermont, Vasington
Dominanta kategorija: Family-friendly
Prosecna cena: 1600 dolara
Prosecna kvadratura: oko 80 kvadrata

Klaster 3 - za svakog po nesto, klasicni¶

Kardinalnost: 899
Teritorija: Vasington, Juta, Kolorado, Masacusets (oko Bostona)
Dominanta kategorija: Turbulent + Nature-lovely
Prosecna cena: oko 1600 dolara
Prosecna kvadratura: oko 88 kvadrata

Klaster 4 - najbolji odnos cene i kvadrature¶

Kardinalnost: 2515 (najbrojniji klaster) Teritorija: Novi Meksiko, Teksas, Luizijana, Oklahoma, Arkansas, Misisipi, Alabama, Dzordzija, Juzna Karolina, Kansas, Misuri, Kentaki, Tenesi, Indijana, Ohajo, Micigen, Pensilvanija, Nevada, Ajdaho, Severna i Juzna Dakota, Dominanta kategorija: Family-friendly
Prosecna cena: oko 1300 dolara
Prosecna kvadratura: oko 95 kvadrata

Klaster 5 - za mlade¶

Kardinalnost: 940
Teritorija: Vest Virdzinija, Njujork, Ilinoj, Viskonsin
Dominanta kategorija: Turbulent
Prosecna cena: oko 1500 dolara
Prosecna kvadratura: oko 83 kvadrata